内存对齐如何工作以引用C ++中的结构？

Question

I made a discovery which I was not aware of when casting a buffer to a struct reference which itself includes a data buffer.

I think this example is best explained with a code example:

struct myStruct {
  int a;
  int b;
  char data[1];
};

#define TB_SIZE 5
char data[TB_SIZE] = {0x1, 0x2, 0x3, 0x4, 0x5};


void f(struct myStruct& dest) {
  dest.a = 33;
  memcpy(dest.data, data, TB_SIZE);
}

int main() {
  char buf[100];
  f(reinterpret_cast<struct myStruct&>(buf[0]));
  struct myStruct castStruct = reinterpret_cast<struct myStruct&>(buf[0]);

  printf("Src Values\n");
  for(int i = 0; i < TB_SIZE; ++i) {
    printf("0x%02x, ", data[i]);
  }
  printf("\nActual\n");
  for(int i = 0; i < TB_SIZE; ++i) {
    printf("0x%02x, ", castStruct.data[i]);
  }


  return 0;
}

The output from this simple program is:

Src Values
0x01, 0x02, 0x03, 0x04, 0x05, 
Actual
0x01, 0x02, 0x03, 0x04, 0x00,

After casting the buffer to a struct reference, when i print the data buffer inside the struct, it seems that we have lost the last value, ie. castStruct[4].

Now I guess this has with alignment issues with structs to do but I am not really sure why. If I instead change all reference to pointers the program works as expected, Also if I just have one integer in myStruct before the data[] field the program also works as expected. So my question is simply, how does memory alignment work for a reference to a struct in C++?

Answer 1

So I think this behaviour is expected - in fact you get more than you should reasonably expect, in this line:

struct myStruct castStruct = reinterpret_cast<struct myStruct&>(buf[0]);

You copy the contents of the nasty buffer into a new struct castStruct - the compiler only guarantees that the actual elements of the struct are copied - specifically it only copies a single character array. What you would expect is to only get the first character (as it happens because of alignment an extra few bytes are also copied). Note that castStruct.data[2] refers to unallocated, undefined memory as you have copied the struct away from your first buffer.